a

r
X
i
v
:
1
0
0
2
.
4
8
5
2
v
2


[
h
e
p
-
t
h
]


3
1

M
a
r

2
0
1
0
General Split Helicity Gluon Tree Amplitudes in Open Twistor String Theory
Louise Dolan
Department of Physics
University of North Carolina, Chapel Hill, NC 27599
Peter Goddard
Institute for Advanced Study
Princeton, NJ 08540, USA
Abstract
We evaluate all split helicity gluon tree amplitudes in open twistor string theory. We show
that these amplitudes satisfy the BCFW recurrence relations restricted to the split helicity
case and, hence, that these amplitudes agree with those of gauge theory. To do this we
make a particular choice of the sextic constraints in the link variables that determine the
poles contributing to the contour integral expression for the amplitudes. Using the residue
theorem to re-express this integral in terms of contributions from poles at rational values of
the link variables, which we determine, we evaluate the amplitudes explicitly, regaining the
gauge theory results of Britto et al. [25].
1 Introduction
In this paper, we extend the techniques introduced in [1] to evaluate explicitly the general
split helicity gluon tree amplitudes in open twistor string theory, establishing the gauge
theory recursion relations [2]-[3] for these twistor string amplitudes,. This approach is based
on the use of the link variables of Arkani-Hamed, Cachazo, Cheung and Kaplan [4, 5]; see
also [6]-[8].
Twistor string theory [9]-[11] provides tree level four-dimensional N = 4 Yang-Mills theory
with a potential string description. Particular examples of the derivation of gauge tree
amplitudes from the twistor string are given in [12] -[22], and recent ones using link variables
are given in [1], [23] and [24]. Gauge eld theory formulae for split helicity trees are found
in [25]-[33].
We consider the twistor string tree amplitude with m positive helicity gluons, labeled
i
1
, . . . , i
m
, and n negative helicity gluons, labeled r
1
, . . . , r
n
, with the helicities of the same
sign being adjacent, i.e. the split helicity tree amplitude. Write P = {i
1
, . . . , i
m
} and
N = {r
1
, . . . , r
n
}; and the gluon momenta p
a
a
=
a

a
. The link variables c
lu
, l P, u N
satisfy the 2(m + n) linear equations

l
=

uN
c
lu

u
(1)

u
=

lP

l
c
lu
. (2)
where we have suppressed the spinor indices. (See [34, 1] for our conventions.) These equa-
tions are not independent because they imply momentum conservation [4], and for momenta
satisfying this consistency condition they provide 2(m + n) 4 constraints on the mn vari-
ables c
lu
, leaving N
R
= (m 2)(n 2) degrees of freedom. Fixing i, j P and r, s N,
we can take the independent degrees of freedom to be c
kt
, where k P

, t N

, where
P

= {k P, k = i, j} and N

= {t N, t = r, s}, the remaining c

lu
being expressed in
terms of these N
R
variables using (1, 2).
In [1] we showed how to write the general twistor string tree amplitude as a contour integral
over the N
R
variables c
kt
, k P

, t N

,
M
mn
= K
mn
_
O
F(c)

kP

tN

dc
kt
C
kt
, (3)
where K
mn
= r, s
2m
[i, j]
2n
and F(c) is a simple rational function of the c
lu
, the form of
which is given by (4.12) of [1], the sextic functions C
kt
C
ijk
rst
are given by the determinants,
C
ijk
rst
=

c
is
c
it
c
it
c
ir
c
ir
c
is
c
js
c
jt
c
jt
c
jr
c
jr
c
js
c
ks
c
kt
c
kt
c
kr
c
kr
c
ks

, (4)
2
and the contour O is chosen so as to include the residue contributions from each of the
simultaneous zeros of the C
kt
, but none of those arising from poles of F(c). The N
R
conditions
C
kt
= 0, k P

, t N

, which can be viewed as constraints on the mn variables c

lu
, are
equivalent to the condition that the mn matrix with entries c
1
lu
, l P, u N, has rank 2,
which in turn is the condition that the link variables c
lu
are of the form implied by twistor
string theory [1]. In the split helicity case, F(c) is given by (4.11) of [1] when the xed labels
i = i
1
, j = i
m
, r = r
1
, s = r
n
.
The key observation in evaluating the split helicity amplitudes is that, in this case, the
integrand of (3) simplies if we replace the N
R
constraints C
kt
= C
ijk
rst
, k P

, t N

, by
another independent set in which the labels are contiguous, namely,
C
ab
= C
i
a1
iai
a+1
r
b1
r
b
r
b+1
, 2 a m1, 2 b n 1, (5)
where C
i
a1
iai
a+1
r
b1
r
b
r
b+1
is dened by an equation of the form (4). Using these contiguous constraints,
M
mn
= K
mn
_
O

F(c)
m1

a=2
n1

b=2
dc
ab
C
ab
, (6)
where c
ab
= c
iar
b
, the contour O is now chosen so as to include the residue contributions
from each of the simultaneous zeros of the C
ab
, but none of those arising from poles of

F(c),
which has the simple form

F(c) =
(c)
c
i
1
i
2
r
1
r
2
c
i
m1
im
r
n1
rn
, (7)
where c
kl
tu
= c
kt
c
lu
c
ku
c
lt
, and (c) is a multinomial expression in the c
ab
that we dene in
section 2. It is this simple form, with only two factors in the denominator, that makes the
evaluation relatively straightforward.
In section 2, we rst calculate the factor J(c) that relates the function F(c), in the integrand
of the amplitude, appropriate to our original choice of constraints [1], to the simpler integrand
function

F(c), appropriate to the choice of contiguous constraints (5); and then, second, we
introduce a parametrization,

ab of the solutions of the linear conditions (1), (2) appropriate
to the contiguous constraints and calculate the Jacobian necessary to write M as an integral
over

ab
,
M
mn
=

K
mn
_
O

F(c)
m1

a=2
n1

b=2
d

ab
C
ab
, (8)

K
mn
=
m2

a=2
[i
a
, i
a+1
]
n2
n2

b=2
r
b
, r
b+1

m2
. (9)
In section 3, we show how the simple form of

F(c) for the general split-helicity tree enables
the amplitude to be written in terms of similar expressions involving fewer integrations using
the global residue theorem. In this way, we inductively express M
mn
in terms of a sum
of terms, corresponding to Young diagrams, given by the residue of the integrand at poles
3
specied by N
R
conditions of the form c
iaia+1
r
b
r
b+1
= 0. In section 4, we describe how to solve
these conditions iteratively to give expressions for c
lu
at the poles.
Using this analysis, we establish the BCFW recursion relation for these amplitudes in section
5, thus demonstrating that the twistor string theory yields the gauge theory tree amplitudes
in the split helicity case. In section 6, we apply the results of section 4 to derive explicit
expressions for the terms contributing to M
mn
, obtaining the gauge theory results of Britto
et al. [25]. In section 7, we use our general expression to evaluate the particular cases of the
(4,4) and (5,3) split helicity trees. Section 8 contains some comments about the extension of
our approach to non-split helicity tree amplitudes.
2 Contiguous Constraints
2.1 The Integrand

F(c)
For split helicity amplitudes, the integrand F(c) in (3) is given by (4.11) in [1],
F(c) =
_
c
ij
rs
_
N
R
+1
c
ir
c
js
m1

a=1
1
c
iai
a+1
rs
n1

b=1
1
c
ij
r
b
r
b+1
m1

a=2
n1

b=2
1
c
iar
b
m1

a=2
c
n2
iar
c
n2
ias
n1

b=2
c
m2
ir
b
c
m2
jr
b
, (10)
where i
1
= i, i
m
= j, r
1
= r, r
n
= s.
Now, since
C
ijk
rst
= c
ij
rs
c
ik
st
c
kr
c
jt
c
ij
st
c
ik
rs
c
kt
c
jr
,
C
ijg
rst
= C
igk
rst
c
ij
rs
c
jt
c
ik
rs
c
kt
, when C
ijk
rst
= 0,
and, so for the purpose of calculating residues at the zeros of constraints, we have the
equivalences
C
ijg
rst
C
ijk
rst
C
igk
rst
C
ijk
rst
c
ij
rs
c
jt
c
ik
rs
c
kt
C
igk
rst
C
jgk
rst
c
ij
rs
c
it
c
jt
c
kg
rs
c
kt
c
gt
.
In this sense the product of constraints

kP

tN

C
kt
=
n2

b=1
m2

a=1
C
i
1
i
a+1
im
r
1
r
b+1
rn
J(c)
m2

a=1
n2

b=1
C
iai
a+1
i
a+2
r
b
r
b+1
r
b+2
, (11)
4
where
J(c) =
n2

b=1
m2

a=2
c
ij
rs
c
ir
b+1
c
jr
b+1
c
iai
a+1
rs
c
iar
b+1
c
i
a+1
r
b+1
m2

a=1
n2

b=2
c
iai
a+1
rs
c
i
a+2
r
c
i
a+2
s
c
iai
a+1
r
b
r
b+1
c
i
a+2
r
b
c
i
a+2
r
b+1
=
n2

b=1
c
ii
2
rs
c
i
2
r
b+1
c
ij
rs
c
jr
b+1
m2

a=1
c
iai
a+1
rr
2
c
i
a+2
r
2
c
iai
a+1
rs
c
i
a+2
s
n2

b=1
m2

a=1
c
ij
rs
c
ir
b+1
c
jr
b+1
c
i
a+2
r
c
i
a+2
s
c
iai
a+1
r
b
r
b+1
c
iar
b+1
c
i
a+1
r
b+1
c
i
a+2
r
b
c
i
a+2
r
b+1
.
Using
c
ii
2
rs
c
i
2
r
b+1
c
ij
rs
c
jr
b+1

c
ii
2
r
b
r
b+1
c
i
2
r
c
i
2
s
c
jr
b
c
ij
r
b
r
b+1
c
jr
c
js
c
i
2
r
b
,
we have
J(c) c
ii
2
rr
2
c
i
m1
j
r
n1
s
_
c
ij
rs
_
N
R
+1
c
ir
c
js
n1

b=1
1
c
ij
r
b
r
b+1
m1

a=1
1
c
iai
a+1
rs
m1

a=2
[c
iar
c
ias
]
n2
n1

b=2
[c
ir
b
c
jr
b
]
m2

n2

b=2
m2

a=2
1
c
iai
a+1
r
b
r
b+1
c
iar
b
c
i
a+1
r
b+1
n2

b=2
1
c
ir
b
c
jr
b+1
m2

a=2
1
c
iar
c
i
a+1
s
n1

b=2
m1

a=2
1
c
iar
b
n1

b=2
m1

a=2
1
c
iar
b
,
F(c)
_
n2

b=1
m2

a=1
C
i
1
i
a+1
im
r
1
r
b+1
rn


F(c)
_
m2

a=1
n2

b=1
C
iai
a+1
i
a+2
r
b
r
b+1
r
b+2
, (12)
where

F(c) =
F(c)
J(c)
, and so that the relevant function for contiguous constraints is

F(c) =
1
c
i
1
i
2
r
1
r
2
c
i
m1
im
r
n1
rn
n2

b=2
m2

a=2
c
iai
a+1
r
b
r
b+1
c
iar
b
c
i
a+1
r
b+1
n2

b=2
c
ir
b
c
jr
b+1
m2

a=2
c
iar
c
i
a+1
s
n1

b=2
m1

a=2
c
iar
b
.
(13)
Writing f
ab
= c
iai
a+1
r
b
r
b+1
, the integrand becomes
I
mn
=

F(c)

n1
b=2

m1
a=2
C
i
a1
iai
a+1
r
b1
r
b
r
b+1
=
1
f
11
f
m1,n1
n2

b=2
m2

a=2
f
ab
c
ab
c
a+1,b+1
n2

b=2
c
1b
c
m,b+1
m2

a=2
c
a1
c
a+1,n
n1

b=2
m1

a=2
c
ab
C
ab
, (14)
and for MHV amplitudes, this gives
I
m2
=
1
c
11
c
m2
m1

a=1
1
f
a1
, I
2n
=
1
c
11
c
2n
n1

b=1
1
f
1b
. (15)
5
For low values of m, n, we have the following:
I
33
=
c
22
f
11
f
22
C
22
; I
43
=
c
21
c
22
c
32
c
33
f
11
f
32
C
22
C
32
; (16)
I
53
=
c
21
c
22
c
31
c
32
c
33
c
42
c
43
f
11
f
42
C
22
C
32
C
42
; I
44
=
f
22
c
12
c
21
c
2
22
c
23
c
32
c
2
33
c
34
c
43
f
11
f
33
C
22
C
23
C
32
C
33
. (17)
2.2 Parameterization of c
lu
In [1] it was shown that the general solution for c
lu
of the linear conditions (1), (2) can be
written in the form
c
lu
= a
lu
+
1
4

g,hP
v,wN

lgh
uvw
[g, h]v, w, (18)
where a
lu
is a particular solution and
lgh
uvw
is antisymmetric under permutations of l, g, h
and also under permutations of u, v, w. (Note this denition of
lgh
uvw
diers by a factor of p
2
from that used in [1].) Because there are only N
R
independent solutions to (1), (2), there is
some arbitrariness in the choice of the parameters
lgh
uvw
; in [1] this arbitrariness was resolved
by requiring that
lgh
uvw
= 0 unless i, j {l, g, h} and r, s {u, v, w}, so that the all the
lgh
uvw
are related to the N
R
parameters

ab
=
iiaj
rr
b
s
, 2 a m1, 2 b n 1,
with i = i
1
, j = i
m
, r = r
1
, s = r
n
.
This choice of parameters is appropriate when using the constraints C
kt
, but, when using
contiguous constraints C
ab
, it is more appropriate to replace
lgh
uvw
with parameters

lgh
uvw
such that

lgh
uvw
= 0 unless {l, g, h} = {i
a1
, i
a
, i
a+1
}, for some a, where 2 a m 1, and
{u, v, w} = {r
b1
, r
b
, r
b+1
}, for some b, where 2 b n 1. Write

ab
=
i
a1
iai
a+1
r
b1
r
b
r
b+1
, 2 a m1, 2 b n 1;
then, for 2 a m1 and 2 b n 1,
c
iar
b
a
iar
b
=
ab
[i, j]r, s
=

=1,0,1
2a+a

=1,0,1
2b+b

n
(1)
a

+b

a+a

,b+b
[i
(a,a

)
, i
(a,a

)
]r
(b,b

)
, r
(b,b

)
,
(19)
where (a, 1) = a 2, (a, 0) = a 1, (a, 1) = a + 1 and (a, 1) = a 1, (a, 0) =
a + 1, (a, 1) = a + 2.
6
From (6),
M
mn
=

K
mn
_
O

F(c)
m1

a=2
n1

b=2
d
ab
C
ab
,

K
mn
= [i, j]
(m3)(n2)
r, s
(m2)(n3)
,
=

K
mn
_
O

F(c)

m1

a=2
n1

b=2
d

ab
C
ab
. (20)
The Jacobian can be computed in two stages. First dene

ab
r, s =

=1,0,1
2b+b

n
(1)
b

a,b+b
r
(b,b

)
, r
(b,b

)
,
then

ab
[i, j] =

=1,0,1
2a+a

m
(1)
a

a+a

,b
[i
(a,a

)
, i
(a,a

)
].
Now if J
m,b
[i, j]
m+2
denotes the Jacobian of
ab
with respect to

ab
, 2 a m 2, J
m,b
satises the recurrence relation
J
m,b
= [i
m2
, i
m
]J
m1,b
[i
m3
, i
m2
][i
m1
, i
m
]J
m2,b
, (21)
which uniquely species J
m,b
subject to the conditions J
3,b
= [i
1
, i
3
], J
4,b
= [i
1
, i
4
][i
2
, i
3
], and
the solution is provided by
J
m,b
= [i
1
, i
m
]
m2

a=2
[i
a
, i
a+1
].
Then, for xed b, the (m2) (m2) Jacobian

=
m2

a=2
[i
a
, i
a+1
]
[i, j]
.
Similarly, for xed a, the (n 2) (n 2) Jacobian

=
n2

b=2
r
b
, r
b+1

r, s
,
so that the full Jacobian

=
_
m2

a=2
[i
a
, i
a+1
]
[i, j]
_
n2
_
n2

b=2
r
b
, r
b+1

r, s
_
m2
, (22)
leading to (8) and (9).
7
3 Towards a Recurrence Relation
Exploiting the simple form of the denominator of

F(c), we can use the global residue theorem
[4, 1] to reduce the number of integrations. First we illustrate this for
M
44
=

K
44
_
O
I
44
d

22
d

23
d

32
d

33
, (23)
where I
44
is given by (17). To evaluate (23) by the global residue theorem, we divide the
factors in the denominator of the integrand into disjoint subsets:

22
= {C
22
};
23
= {C
23
, f
11
, f
33
};
32
= {C
32
};
33
= {C
33
},
M
4,4
= R
_
C
22
C
23
C
32
C
33
_
= R
_
C
22
f
11
C
32
C
33
_
R
_
C
22
f
33
C
32
C
33
_
, (24)
where we are using the matrix notation
R
_
C
22
C
23
C
32
C
33
_
for the residue that would be denoted by R(C
22
, C
23
, C
32
, C
33
) in the notation of [1].
Since C
22
= f
11
f
22
c
13
c
31
f
12
f
21
c
11
c
33
, then for f
11
= 0, C
22
= f
12
f
21
c
11
c
33
; but c
33
is
in the numerator of the integrand and when f
11
= 0, c
11
c
22
= c
12
c
21
, which is also in the
numerator, so that, when f
11
= 0, the relevant zeros of C
22
occur at f
12
= 0 and at f
21
= 0.
Thus
R
_
C
22
f
11
C
32
C
33
_
= R
_
f
12
f
11
C
32
C
33
_
+R
_
f
21
f
11
C
32
C
33
_
. (25)
Further, since C
32
= f
21
f
32
c
23
c
41
f
22
f
31
c
21
c
43
, then for f
21
= 0, C
32
= f
22
f
31
c
21
c
43
and
so the integrand for the last term in (25) is, up to a constant
c
3
22
c
23
c
32
c
33
c
34
f
11
f
33
f
12
f
21
C
23
f
31
c
21
C
33
, (26)
so that we do not have to consider the zeros of C
32
corresponding to f
22
= 0 or c
43
= 0.
The zero of C
32
corresponding to c
21
= 0 when f
11
= f
21
= 0, entails c
11
c
22
= c
31
c
22
= 0;
the presence of c
22
in the numerator, means that we would need c
11
= c
21
= c
31
= 0 for a
nonzero residue. But, from (2), this would imply that
i
4
is parallel to
r
1
, and so for generic
momenta there is no contribution from c
21
= 0.
This leaves us to consider the possibility of a contribution to the last term in (25) from
f
11
= f
21
= f
31
= 0. Then
c
32
(
r
2
c
21

r
1
c
22
) = c
32

i
4
(c
22
c
41
c
21
c
42
) = 0. (27)
Since c
32
is in the numerator of (26), it must be nonzero for a nonzero contribution and then
8
(27) implies that
r
1
and
r
2
are parallel. So the last term in (25) vanishes, leaving
R
_
C
22
f
11
C
32
C
33
_
= R
_
f
12
f
11
C
32
C
33
_
. (28)
Similarly
R
_
C
22
f
33
C
32
C
33
_
= R
_
C
22
f
33
C
32
f
23
_
, (29)
so that (24) becomes
M
44
= R
_
C
22
C
23
C
32
C
33
_
= R
_
f
11
f
12
C
32
C
33
_
+R
_
C
22
f
23
C
32
f
33
_
, (30)
using the antisymmetry of the residue R on interchanging its arguments. The integrands of
the terms on the right hand side of (30) are:
for R
_
f
11
f
12
C
32
C
33
_
,
c
21
c
2
22
c
23
c
32
c
33
c
43
f
11
f
12
c
11
f
13
f
21
f
33
C
32
C
33
=
1
f
11
f
12
I
34
c
21
c
22
c
23
f
13
c
11
, (31)
where I
34
by dened from (14) with the i
a
shifted to i
a+1
, and
for R
_
C
22
f
23
C
32
f
33
_
,
c
21
c
22
c
23
c
32
c
2
33
c
43
f
23
f
33
c
44
f
11
f
13
f
32
C
22
C
32
=
1
f
23
f
33
I
43
c
23
c
33
c
43
f
13
c
44
. (32)
We shall see in section A that the factors c
21
c
22
c
23
/f
13
c
11
and c
23
c
33
c
43
/f
13
c
44
in (31) and
(32), respectively, correspond to appropriate Jacobians, so that (30) expresses M
44
in terms
of M
34
and M
43
.
In a similar fashion we can establish a relation for the general (m, n) split helicity amplitude,
M
m,n
= R
_
_
_
_
_
_
_
C
22
C
23
. . . C
2,n2
C
2,n1
C
32
C
33
. . . C
3,n2
C
3,n1
.
.
.
.
.
.
.
.
.
.
.
.
C
m2,2
C
m2,3
. . . C
m2,n2
C
m2,n1
C
m1,2
C
m1,3
. . . C
m1,n2
C
m1,n1
_
_
_
_
_
_
_
, (33)
whose integrand is given by (14).
9
As in (24), we have
M
m,n
= R
_
_
_
_
_
_
_
C
22
C
23
. . . C
2,n2
f
11
C
32
C
33
. . . C
3,n2
C
3,n1
.
.
.
.
.
.
.
.
.
.
.
.
C
m2,2
C
m2,3
. . . C
m2,n2
C
m2,n1
C
m1,2
C
m1,3
. . . C
m1,n2
C
m1,n1
_
_
_
_
_
_
_
R
_
_
_
_
_
_
_
C
22
C
23
. . . C
2,n2
f
m1,n1
C
32
C
33
. . . C
3,n2
C
3,n1
.
.
.
.
.
.
.
.
.
.
.
.
C
m2,2
C
m2,3
. . . C
m2,n2
C
m2,n1
C
m1,2
C
m1,3
. . . C
m1,n2
C
m1,n1
_
_
_
_
_
_
_
. (34)
Suppose rst that m, n 4. When f
11
= 0, C
22
= f
12
f
21
c
11
c
33
, for the same reason as
in the case m = n = 4 just discussed, the relevant zeros of C
22
occur at f
12
= 0 and at
f
21
= 0, so that the rst term on the right hand side of (34) can be written as the sum of two
terms replacing C
22
by f
12
and f
21
, respectively. In the case f
21
= 0, C
32
= f
22
f
31
c
21
c
43
;
f
22
, c
43
are in the numerator and, if m > 4, c
21
is as well, and the only zero that needs to be
considered is those of f
31
. We can then proceed inductively: if f
a1
= 0, with 1 a m2,
then C
a+1,2
= f
a2
f
a+1,1
c
a1
c
a+2,3
, and f
a2
, c
a+2,3
are in the numerator and c
a1
is as well if
1 a < m2, so that the relevant zero comes from f
a+1,1
= 0. When we eventually reach
a = m2 (as we do immediately if m = 4), c
m2,1
is not in the numerator; but if c
m2,1
= 0
and f
b1
= 0, 1 b m 2, we are led to c
1b
= 0, 1 b m 1, as in the m = n = 4 case,
which is excluded because it would imply that
im
is parallel to
r
1
. So for the rst term on
the right hand side of (34) to be nonzero we need f
b1
= 0, 1 b m1, but then a similar
argument to that used in (27) would imply that
r
1
and
r
2
are parallel, so excluding this
possibility.
It follows that, in the rst term on the right hand side of (34), we only need to consider the
zeros of C
22
corresponding to f
12
= 0. In this case, arguments like those just applied to the
constraint functions in the rst column can now be applied along the rst column, to deduce
that the relevant zeros of C
2b
come from f
1b
= 0, 2 b n 2. The dierence is that the
nal constraint C
2,n1
is absent, so in this instance we are not led to infer
i
1
and
i
2
are
parallel, which would have excluded this contribution as well. Exactly similar arguments
applied to the second term on the right hand side of (34) lead to the conclusion that the
corresponding contributions come from replacing C
a+1,n1
by f
a,n1
, 2 a m2, so that
10
(34) becomes
M
mn
= (1)
n
R
_
_
_
_
_
_
_
f
11
f
12
. . . f
1,n3
f
1,n2
C
32
C
33
. . . C
3,n2
C
3,n1
.
.
.
.
.
.
.
.
.
.
.
.
C
m2,2
C
m2,3
. . . C
m2,n2
C
m2,n1
C
m1,2
C
m1,3
. . . C
m1,n2
C
m1,n1
_
_
_
_
_
_
_
+ (1)
m
R
_
_
_
_
_
_
_
C
22
C
23
. . . C
2,n2
f
2,n1
C
32
C
33
. . . C
3,n2
f
3,n1
.
.
.
.
.
.
.
.
.
.
.
.
C
m2,2
C
m2,3
. . . C
m2,n2
f
m2,n1
C
m1,2
C
m1,3
. . . C
m1,n2
f
m1,n1
_
_
_
_
_
_
_
, (35)
where we have again used the antisymmetry of the residue.
It is straightforward to establish this relation also for m = 3,
M
3n
= (1)
n
R(f
11
, f
12
, . . . , f
1,n3
, f
1,n2
) R(C
22
, C
23
, . . . , C
2,n2
, f
2,n1
),
using similar arguments, and likewise for n = 3.
Generalizing (31) and (32), the two terms in the relation (35) for M
mn
have integrands
(1)
n
_
n2

b=1
1
f
1b
_
I
m1,n
1
c
11
f
1,n1
n1

b=1
c
2b
(36)
and
(1)
m
_
m1

a=2
1
f
a,n1
_
I
m,n1
1
c
mn
f
1,n1
m

a=2
c
a,n1
, (37)
respectively. Again, we shall see in section A that the nal factors in (36) and (37) are
appropriate Jacobians.
By applying (35) iteratively we can express M
mn
in terms of a sum of contributions from
poles all specied by conditions of the form
f
ab
c
ab
c
a+1,b+1
c
a,b+1
c
a+1,b
= 0. (38)
The iterative process has to be continued with the various terms generated until each term
involves I
m

n
with either m

= 2 or n

= 2, at which point all the constraint functions C

ab
have been removed and each term must involve N
R
= (m2)(n 2) conditions of the form
(38).
The various terms are distinguished by the particular set of N
R
conditions (38). Not all sets
of N
R
conditions of this form are permitted. To characterize the particular set of conditions
associated with a particular term in the expression for M
mn
, we start with a (m2)(n2)
matrix or grid, which we label by pairs U = {(a, b) : 2 a m 1, 2 b n 1}. The
11
rst term in (35), i.e. (36), corresponds to applying the conditions f
1,b1
= 0, 2 b n1,
which we can associate with the top row of our grid, and second term in (35), i.e. (37),
corresponds to applying the conditions f
a,n1
= 0, 2 a m 1, which we can associate
with the right hand column of our grid. We denote the rst by placing the symbol

in each
of the places or squares on the rst row of the grid and the second by placing the symbol
in each of the squares on the last column. As we iteratively apply the relation (35), we ll
either the top row or the last column of the grid formed by unlled squares. In this way we
obtain a diagram of the form shown in Figure 1.

Figure 1. A Young-like Diagram
Generically, in such a diagram the

may or may not stretch all the way across the top row,
and may or may not stretch all the way down the rst column, starting from the top left
(2, 2). Similarly, the may or may not stretch all the way up the last column, and may
or may not stretch all the way back along the bottom row starting from (m 1, n 1).
However, either the

must stretch all the way across the top row or the must stretch all
the way up the last column. More generally, if V denotes the subset of the grid U occupied
by

, the constraint on the allowable sets of conditions is that if (a, b) V then (c, d) V
whenever c a and d b. Similarly, if W denotes the complementary subset of U to V, i.e.
the squares occupied by , then if (a, b) W and a c and b d then (c, d) W. The set
of conditions characterizing the term are then
f
a1,b1
= 0, (a, b) V, f
ab
= 0, (a, b) W. (39)
The conditions on the grid ensure that it has the form of a set of steps, i.e. the form of a
Young diagram [36]. The number of such diagrams is
(m + n 4)!
(m2)!(n 2)!
, (40)
so that this is the number of terms in the expression for M
mn
. The diagrams obtained in
this way are in correspondence with the zigzag diagrams introduced by Britto et al. [25] for
the gauge theory, as would be expected.
In some ways it is preferable to replace the diagrams of the form of Figure 1 by equivalent
ones drawn in an mn grid: we draw a box linking the centers of the squares (a, b), (a, b +
12
1), (a + 1, b), (a + 1, b + 1) if f
ab
= 0 is one of our conditions, which is equivalent to having
(a + 1, b + 1) V or (a, b) W. Then, corresponding to Figure 1, we have the alternative
representation shown in Figure 2, where we have labelled the rows and the columns by the
corresponding positive and negative helicities respectively.
This step diagram has the property that if the squares (a, b), (a, d), (c, b), (c, d) are linked by
a box then c
ac
bd
= c
ab
c
cd
c
ad
c
bc
= 0. [It is possible to have c
ac
bd
= 0 and c
ac
de
= 0 without
having c
ac
be
= 0 because the rst two equations will hold if c
ad
= c
cd
= 0, but we exclude such
congurations here because the factors in the numerator of our integrands mean that they
are not relevant.]
i
1
i
2
i
3
i
4
i
5
i
6
i
7
i
8
i
9
r
1
r
2
r
3
r
4
r
5
r
6
r
7
r
8
r
9
r
10
r
11
Figure 2. A Step Diagram
As an example, for (m, n) = (4, 4), we have 6 possible diagrams corresponding to the six
terms in the known representations of the (4, 4) split helicity amplitude [13, 2]; these are:
Figure 3. Step Diagrams for the (4, 4) Split Helicity Amplitude.
To turn the relation (35) into a recurrence relation that expresses M
mn
in terms of M
m1,n
and M
m,n1
we need to show how the appropriate set of conditions f
ab
= 0 enable the
13
reduction of the m + n equations (1) and (2) for the mn variables c
lu
, l P, u N, to a
similar m + n 1 equations for a reduced set of variables, and, in so doing, to specify the
m + n 1 momenta that are the arguments of reduced amplitudes M
m1,n
and M
m,n1
.
We shall explain how to do this in the next section. Then we need to evaluate the Jacobians
involved in the calculation of the residues in (35) and in relating the integration variables
that remain after taking the residue to those appropriate to M
m1,n
and M
m,n1
. This we
do in section A.
4 Reducing the Equations for c
lu
For a specic step diagram, the conditions (39), together with (1) and (2), enable us to
evaluate the link variables c
lu
iteratively.
(a) Suppose that the step diagram has m
1
1 rows of maximal length, i.e. n 2 boxes, and
that m
1
> 1. Let S = {i
2
, . . . , i
m
1
} and write i = i
1
, j = i
m
, r = r
1
, s = r
n
as usual. Then,
for this diagram,
c
ik
tu
= 0, k S, t, u N, t, u = s, (41)
and, from (1),
c
iu

k
c
ku

i
=

tN
c
ik
ut

t
= c
ik
us

s
, k S, u N, u = s. (42)
Using these equations, we can express c
ku
in terms of c
iu
, k S, u N:
c
ku
k, s
=
c
iu
i, s
, k S, u N, u = s, (43)
c
ks
k, s
=
c
is
i, s

k, i
k, si, s
, k S. (44)
Then, for t N, t = s, we can write (2)

t
=
_
_

kS

k
k, s
i, s
_
_
c
it
+

kPS

k
c
kt
,
=

i
c
it
+

kPS

k
c
kt
, (45)
where S = S {i, s} and the shifted

i
is dened as

i
=

kS

k
k, s
i, s

1
i, s
P
S

s
; (46)
14
further

s
=

kP

k
c
ks
=
_
_

kS

k
k, s
i, s
_
_
c
is
+
_

kS

k
k, i
s, i
_
+

kPS

k
c
ks
,

s
=

i
c
is
+

kPS

k
c
ks
,
(47)
with

s
=

kS

k
k, i
s, i

1
s, i
P
S

i
. (48)
So taking the m

= m|S| = mm
1
+1 equations (1) for k P S and the new n equations
(45) and (47), we have a smaller system of equations with (m, n) replaced by (m

, n), that is
we have reduced the number of equations by m
1
1.
Further, we note that

i
(

i
)
T
+
s
(

s
)
T
=

kS
_

s
k, i
s, i
+
i
k, s
i, s
_
(
k
)
T
=

kS

k
(
k
)
T
P
S
, (49)
as can be veried by taking the angle bracket with
i
and
s
. Squaring this equation,
i, s[i

, s

] =
_
p
i
+

kS
p
k
+ p
s
_
2
= p
2
S
. (50)
(b) If the step diagram has no rows of maximal length (i.e. m
1
= 1), it must have some
number of columns of maximal height, say n n
1
, n > n
1
. Let T = {r
n
1
, . . . , r
n1
} Then,
for this diagram,
c
kl
ts
= 0, t T , k, l P, k, l = i, (51)
and, from (2),

t
c
ks

s
c
kt
=

lP

l
c
kl
ts
=
i
c
ki
ts
, t T , k P, k = i. (52)
From these relations, as in (a), we can replace (1), (2) with a reduced set of equations

k
=

tNT
c
kt

t
+ c
ks

s
, k P, k = i, (53)

i
=

tNT
c
it

t
+ c
is

s
, (54)
15
where T = T {i, s} and the shifted

s
and

i
are dened by

s
=

tT

t
[t, i]
[s, i]

1
[s, i]
P
T

i
, (55)

i
=

tT

t
[t, s]
[i, s]

1
[i, s]
P
T

s
. (56)
So taking the n

= n |T | = n
1
equations (2) for t N T and the new m equations (53)
and (54), we have a smaller system of equations with (m, n) replaced by (m, n
1
), that is we
have reduced the number of equations by n n
1
.
As before, in (49) and (50),

T
i
+

T
s
=

tT

t
(
t
)
T
P
T
, (57)
and
i

, s

[i, s] =
_
p
i
+

tT
p
t
+ p
s
_
2
= p
2
T
. (58)
5 The Recurrence Relation
In (35), M
mn
is written as the sum of two contributions and, from (36), the rst of which is

K
mn
_

F
m1,n
(c)
c
11
f
1,n1

1 n1

b=1
c
2b
m1

a=3
n1

b=2
d

ab
C
ab
. (59)
Using (88), this becomes
r
n
, i
2

n1
[i
2
, i
3
]
n2
r
n
|P
i
1
i
2
|i
3
]
n2
i
1
, i
2
r
n
, i
1

K
m1,n
_

F
m1,n
(c)
m1

a=3
n1

b=2
d

ab
C
ab
(60)
If we use the conditions f
1b
= 0, 1 b n 2, to eliminate
i
1
,
i
1
, c
1b
, 1 b n, to dene
a new system as in section 4(a), consisting of

i
2
,

i
2
=
1
i
2
, r
n

P
S

rn
;
ia
,
ia
, 3 a m; (61)

r
b
,
r
b
, 1 b n 1;
rn
,

rn
=
1
r
n
, i
2

P
S

i
2
; (62)
where S = {i
1
, i
2
, r
n
}, and c
ab
, 2 a m, 1 b n, and then use this system to dene a
set of contiguous s,

ab
, 3 a m 1, 2 b n 1, these will not be the same as the
corresponding

ab
. So we have to incorporate the corresponding Jacobian (91)
16
r
n
, i
2

n1
[i
2
, i
3
]
n2
r
n
|P
i
1
i
2
|i
3
]
n2
i
1
, i
2
r
n
, i
1

K
m1,n
_

F
m1,n
(c)

1
m1

a=3
n1

b=2
d

ab
C
ab
=
r
n
, i
2

r
n
, i
1
i
1
, i
2

K
m1,n
_

F
m1,n
(c)
m1

a=3
n1

b=2
d

ab
C
ab
=
r
n
, i
2

r
n
, i
1
i
1
, i
2

M
m1,n
. (63)
Making a similar evaluation of the second contribution M
mn
in (35), we have
M
mn
=
r
n
, i
2

r
n
, i
1
i
1
, i
2

M
m1,n
+
[r
n1
, i
1
]
[r
n1
, r
n
][r
n
, i
1
]
M
m,n1
. (64)
Note this is the BCFW recursion relation restricted to split helicity amplitudes [2, 3], where
the shifted momenta

i
,

s
in M
m1,n
are given by (46) and (48), and the shifted momenta

i
,

s
in M
m,n1
are given by (56) and (55), correspond to the BCFW reference momenta.
So we have a complete proof that the split amplitudes for the twistor string agree with gauge
theory.
Note, from (15),
M
m2
=
K
m2
c
11
c
m2
m1

a=1
1
f
a1
=
r
1
, r
2

3
r
2
, i
1
i
m
, r
1

m1

a=1
1
i
a
, i
a+1

(65)
M
2n
=
K
2n
c
11
c
2n
n1

b=1
1
f
1b
=
[i
1
, i
2
]
3
[r
n
, i
1
][i
2
, r
1
]
n1

b=1
1
[r
b
, r
b+1
]
, (66)
using
c
a1
=
i
a
, r
2

r
1
, r
2

, c
a2
=
i
a
, r
1

r
1
, r
2

, f
a1
=
i
a
, i
a+1

r
1
, r
2

,
from (1) in (65), and
c
1b
=
[r
b
, i
2
]
[i
1
, i
2
]
, c
2b
=
[r
b
, i
1
]
[i
1
, i
2
]
, f
1b
=
[r
b
, r
b+1
]
[i
1
, i
2
]
,
from (2) in (66).
We consider applying the recurrence relation times and we calculate the term that comes
from removing the top row each time. This involves M
m,n
with momenta

i
+1
, . . . ,
im
;
r
1
, . . . ,
rn
;

i
+1
,
i
+2
, . . . ,
im
;
r
1
, . . . ,
(+1)
rn
, (67)
17
where

i
1

i
1
and
(1)
rn
=
rn
, and the iterative relation

i
+1
=

, r
n

i
+1
, r
n

+
i
+1
,
(+1)
rn
=

, i
+1

i
+1
, r
n

()
rn
(68)
has the solution

=
1
i

, r
n

P
S

rn
,
()
rn
=
1
r
n
, i

P
S

, (69)
where S

= {r
n
, i
a
: 1 a }. Thus we may build up S

adding one particle at a time.

and it follows that the contribution to M
mn
with f
ab
= 0, 1 a , 1 b n 2, is
r
n
, i
+1

r
n
, i
1

a=1
1
i
a
, i
a+1

M
m,n
. (70)
Similarly the contribution to M
mn
with f
ab
= 0, 2 a m1, n b n 1, is
[r
n
, i
1
]
[r
n
, i
1
]
n1

b=n
1
[r
b
, r
b+1
]
M
m,n
, (71)
where M
m,n
with momenta

(+1)
i
1
, . . . ,
im
;
r
1
, . . . ,

r
n
;
i
1
,
i
+2
, . . . ,
im
;
r
1
, . . . ,
r
n
, (72)

(+1)
i
1
=
1
[i
1
, r
n
]
P
T
+1

r
n
,

r
n
=
1
[r
n
, i
1
]
P
T
+1

i
1
, (73)
and T

= {r
b
, i
1
: n < b n}.
In the next section we extend these results to obtain the expression for the contribution
associated with a general step diagram.
6 General Formula.
In this section we evaluate the contribution to the (m, n) split helicity amplitude M
mn
associated with a general step diagram using the analysis of the previous sections. We can
describe such a diagram as follows: suppose it has m
1
1 rows of maximal length, i.e. n2,
(so that m
1
= 1 if there are no such rows); n n
1
columns of the maximal permitted height
given m
1
, i.e. mm
1
1; m
2
m
1
rows of the next maximal length given n
1
, i.e. n
1
2;
n
1
n
2
columns of height mm
2
1; and so on until we reach n
p1
n
p
columns of height
mm
p
1 and, nally, mm
p
1 rows of length n
p
2.
Associated with this step diagram we have the conditions:
f
ab
= 0, m
q1
a < m
q
, 1 b < n
q1
1, 1 q p + 1, (74)
f
ab
= 0, m
q
< a < m, n
q
b < n
q1
, 1 q p, (75)
18
where m
0
= 1, n
0
= n, m
p+1
= m1.
We evaluate the contribution to M
mn
by successively reducing the amplitude using (70) and
(71). This process is illustrated in Figure 4 for the step diagram specied by
m = 6, n = 8, p = 2, m
1
= 2, m
2
= 3, m
3
= 5, n
1
= 6, n
2
= 3.

Figure 4. Reduction of a Step Diagram

From (70), the contribution to M
m,n
for
f
ab
= 0, 1 a < m
1
, 1 b < n 1,
is
r
n
, i
m
1

r
n
, i
1
i
1
, i
2

m
1
1

a=2
1
i
a
, i
a+1

M
mm
1
+1,n
(i

m
1
, i
m
1
+1
, . . . i
m
, r
1
, . . . , r
n1
, r

n
),
with
[i

m
1
] =
1
i
m
1
, r
n

P
S
1
|r
n
, |i

m
1
= |i
m
1
, |r

n
] =
1
r
n
, i
m
1

P
S
1
|i
m
1
, |r

n
= |r
n
,
S
1
S
m
1
= {r
n
, i
1
, . . . , i
m
1
1
, i
m
1
}, thus eliminating i
1
, . . . , i
m
1
1
.
From (71), the contribution to M
mm
1
+1,n
(i

m
1
, i
m
1
+1
, . . . i
m
, r
1
, . . . , r
n1
, r

n
) for
f
ab
= 0, m
1
< a < m, n
1
b < n,
is
[i

m
1
, r
n
1
]
[i

m
1
, r

n
][r

n
, r
n1
]
n1

b=n
1
+1
1
[r
b
, r
b1
]
M
mm
1
+1,n
1
(i

m
1
, i
m
1
+1
, . . . , i
m
, r
1
, . . . , r
n
1
1
, r

n
1
),
with
|i

m
1
=
1
[i

m
1
, r
n
1
]
P
T
1
|r
n
1
, |i

m
1
] = |i

m
1
], |r

n
1
=
1
[r
n
1
, i

m
1
]
P
T
1
|i

m
1
], |r

n
1
] = |r
n
1
],
T
1
T
n
1
= {r
n
1
, r
n
1
+1
, . . . , r

n
, i

m
1
}, thus eliminating r
n
1
+1
, . . . , r

n
.
Proceeding inductively, from (70), the contribution to
M
mmq+1,nq
(i

mq
, i
mq+1
, . . . , i
m
, r
q
, . . . , r
nq1
, r

nq
)
19
for
f
ab
= 0, m
q
a < m
q+1
, 1 b < n
q
1,
is
r

nq
, i
m
q+1

nq
, i

mq
i

mq
, i
mq+1

m
q+1
1

a=mq+1
1
i
a
, i
a+1

M
mm
q+1
+1,nq
(i

m
q+1
, i
m
q+1
+1
, . . . , i
m
, r
q
, . . . , r
nq1
, r

nq
),
with
[i

m
q+1
] =
1
i
m
q+1
, r

nq

P
S
q+1
|r

nq
, |i

m
q+1
= |i
m
q+1
,
|r

nq
] =
1
r

nq
, i
m
q+1

P
S
q+1
|i
m
q+1
, |r

nq
= |r

nq
,
S
q+1
S
m
q+1
= {r

nq
, i

mq
, . . . , i
m
q+1
1
, i
m
q+1
}, thus eliminating i

mq
, . . . , i
m
q+1
1
.
Proceeding inductively, from (71), the contribution to
M
mm
q+1
+1,nq
(i

m
q+1
, i
m
q+1
+1
, . . . , i
m
, r
1
, . . . , r
nq1
, r

nq
)
for
f
ab
= 0, m
q+1
< a < m, n
q+1
b < n
q
,
is
[i

m
q+1
r
n
q+1
]
[i

m
q+1
r

nq
][r

nq
r
nq1
]
nq1

b=n
q+1
+1
1
[r
b
r
b1
]
M
mmq+1,n
q+1
(i

m
q+1
, i
m
q+1
+1
, . . . , i
m
, r
1
, . . . , r
n
q+1
1
, r

n
q+1
),
with
|i

m
q+1
=
1
[i

m
q+1
, r
n
q+1
]
P
T
q+1
|r
n
q+1
, |i

m
q+1
] = |i

m
q+1
]
|r

n
q+1
=
1
[r
n
q+1
, i

m
q+1
]
P
T
q+1
|i

m
q+1
], |r

n
q+1
] = |r
n
q+1
],
T
q+1
T
n
q+1
= {r
n
q+1
, r
n
q+1
+1
, . . . , r

nq
, i

m
q+1
}, thus eliminating r
n
q+1
+1
, . . . , r

nq
.
It follows that
|i

mq
=
1
[i

mq
, r
nq
]
P
T q
| r
nq
], |i

mq
= |i
mq
,
|i

mq
] = |i

mq
] =
1
i
mq
, r

n
q1
[r
n
q1
,i

m
q1
] . . . i
m
1
, r
n

P
Sq
P
T
q1
. . . P
S
1
| r
n
, (76)
|r

nq
] =
1
r

nq
, i
m
q+1

P
S
1
|i
m
q+1
, |r

nq
] = |r
nq
],
|r

nq
= |r

nq
=
1
[r
nq
, i

mq
]i
mq
, r

n
q1
. . . i
m
1
, r
n

P
T q
P
Sq
. . . P
S
1
|r
n
. (77)
20
So the factor associated with S
q
is
[i

m
q1
, r
n
q1
]
2
r

n
q1
, i
mq

p
2
T
q1
i
m
q1
+1
|P
T
q1
| r
n
q1
]
mq1

a=m
q1
+1
1
i
a
, i
a+1

and the factor associated with T

q
is
r

n
q1
, i
mq

2
[i

mq
, r
nq
]
p
2
Sq
[r
n
q1
1
|P
S
q
|i
mq

n
q1
1

b=nq+1
1
[r
b
, r
b1
]
.
Combining these factors for 1 q p, we obtain
p

q=1
_
_
[i

m
q1
, r
n
q1
]
2
r

n
q1
, i
mq
r

n
q1
, i
mq

2
[i

mq
, r
nq
]
p
2
T
q1
i
m
q1
+1
|P
T
q1
| r
n
q1
]p
2
Sq
[r
n
q1
1
|P
Sq
|i
mq

mq1

a=m
q1
+1
1
i
a
, i
a+1

n
q1
1

b=nq+1
1
[r
b
, r
b1
]
_
_
M
mmp+1,np
(i

mp
, i
mp+1
, . . . , i
m
r
1
, . . . , r
np1
, r

np
),
(78)
We nd the contribution to M
mmp+1,np
in (78) for
f
ab
= 0, m
p
a < m
p+1
m1, 1 b < n
p
1,
is
r

np
i
m1

np
i

mp
i

mp
i
mp+1

m2

a=mp+1
1
i
a
i
a+1

M
2,np
(i

m1
, i
m
, r
1
, . . . , r
np1
, r

np
)
=
r

np
i
m1

np
i

mp
i

mp
i
mp+1

m2

a=mp+1
1
i
a
i
a+1

[i

m1
, i
m
]
3
[i

m1
, r

np
][r

np
, r
np1
][r
1
, i
m
]
np1

b=2
1
[r
b
, r
b1
]
=
[i

mp
, r
np
]
2
r

np
, i
m1

p
2
T p
i
mp+1
|P
T p
| r
np
]
m2

a=mp+1
1
i
a
, i
a+1

np
, i
m1

2
[i

m1
, i
m
]
3
p
2
S
p+1
[r
np1
|P
S
p+1
|i
m1
[r
1
, i
m
]
np1

b=2
1
[r
b
, r
b1
]
,
(79)
where
|i

m1
] =
1
i
m
p+1
, r

np
[r
np
, i

mp
]i
mp
, r

n
p1
[r
n
p1
, i

m
p1
] . . . i
m
1
, r
n

P
S
p+1
P
T p
. . . P
S
1
| r
n
.
Let A = {m
a
: 1 a p}, B = {n
b
: 1 b p}. Combining (78) with (79), we have the
21
general formula for the contribution to M
mn
from the step diagram that we have considered:
1
[r
1
, i
m
]r
n
, i
1

1am2
a/ A
1
i
a
, i
a+1

2bn1
b/ B
1
[r
b
, r
b1
]
p

q=1
1
p
2
T q
i
mq+1
|P
T q
| r
nq
]

p+1

q=1
1
p
2
Sq
[r
n
q1
1
|P
Sq
|i
mq

[i
m
|P
S
p+1
P
T p
. . . P
S
1
| r
n

3
. (80)
This agrees with the gauge theory expression obtained in [25].
Note that P
Sq
= P
S

q
and P
T q
= P
T

q
, where
S

q
= {i
mq+1
, . . . , i
m
, r
1
, . . . , r
n
q1
1
}, T

q
= {i
mq+1
, . . . , i
m
, r
1
, . . . , r
nq1
}.
[Although (80) directly addresses the case where the reduction process both begins and ends
with an S, we can also obtain from the other cases in which the reduction either begins or
ends with a T , or both, by taking m
1
= 1 and S
1
= {r
n
, i
1
} if the process begins with T
and n
p
= 2 and S
p+1
= {i
m
, r
1
} if it ends with T .]
7 Examples
We compute the (4, 4) and (5, 3) amplitudes from twistor string theory using (80).
For (4, 4), the six possible step diagrams are drawn in Figure 3.
a) The lower left diagram corresponds to f
22
= f
23
= f
32
= f
33
= 0. In the notation of
section 6, this diagram has p = 1, n
0
= n = 4, m
1
= 1, n
1
= 2, m
2
= 3 and S
1
= {r
4
, i
1
},
S
2
= {i
4
, r
1
}, T
1
= {r
2
, r
3
, r

4
, i

1
}, where P
T
1
= P
T

1
with T

1
= {i
2
, i
3
, i
4
, r
1
}.
Then from (80) the contribution to M
4,4
is
1
[r
1
i
4
]r
4
i
1

1
i
2
i
3
[r
3
r
2
]
[i
4
|P
i
4
r
1
P
i
2
i
3
i
4
r
1
P
i
4
r
1
|r
4

3
p
2
i
2
i
3
i
4
r
1
i
2
|P
i
2
i
3
i
4
r
1
|r
2
] p
2
i
1
r
4
[r
3
|P
r
4
i
1
|i
1
p
2
i
4
r
1
[r
1
|P
i
4
r
1
|i
3

=
r
1
|P
i
2
i
3
i
4
|i
1
]
3
p
2
i
2
i
3
i
4
r
1
[r
2
|P
i
3
i
4
r
1
|i
2
[r
3
r
4
][r
4
i
1
][r
2
r
3
]i
2
i
3
i
3
i
4
i
4
r
1

.
b) The lower right diagram in Figure 3 has f
11
= f
12
= f
32
= f
33
= 0. This corresponds
to p = 1, n
0
= n = 4, m
1
= 2, n
1
= 2, m
2
= 3 and S
1
= {r
4
, i
1
, i
2
}, S
2
= {i
4
, r
1
},
T
1
= {r
2
, r
3
, r

4
, i

2
}, where P
T
1
= P
T

1
with T

1
= {i
3
, i
4
, r
1
}, from formulae in section 6.
The contribution to M
4,4
yields the second term listed in (81).
c) The lower middle diagram in Figure 3 has f
11
= f
23
= f
32
= f
33
= 0, which is described by
p = 2, n
0
= n = 4, m
1
= 1, n
1
= 3, m
2
= 2 and S
1
= {r
4
, i
1
}, S
2
= {r

3
, i

1
, i
2
}, S
3
= {r
4
, i
1
},
22
T
1
= {r
3
, r

4
, i

1
}, T
2
= {r
2
, r

3
, i

2
}, where P
S
2
= P
S

2
with S

2
= {i
3
, i
4
, r
1
, r
2
}; P
T
1
= P
T

1
with T

1
= {i
2
, i
3
, i
4
, r
1
, r
2
}; P
T
2
= P
T

1
with T

1
= {i
3
, i
4
, r
1
}. From (80), the contribution
to M
4,4
becomes the third term in (81).
The remaining three diagrams are related to the ones directly above them in the Figure 3,
by the symmetry ip i
1
r
4
, i
2
r
3
, i
3
i
2
, i
4
r
1
. Thus their contribution to M
4,4
is found by the ip of the rst three terms, (while exchanging the angle with the square
brackets).
So we have
M
4,4
=
r
1
|P
i
2
i
3
i
4
|i
1
]
3
p
2
i
2
i
3
i
4
r
1
[r
2
|P
i
3
i
4
r
1
|i
2
[r
3
r
4
][r
4
i
1
][r
2
r
3
]i
2
i
3
i
3
i
4
i
4
r
1

+
r
1
|P
i
3
i
4
P
i
1
i
2
|r
4

3
p
2
i
3
i
4
r
1
p
2
r
4
i
1
i
2
[r
2
|P
i
4
r
1
|i
3
[r
3
|P
r
4
i
1
|i
2
[r
2
r
3
]r
1
i
4
i
4
i
3
i
2
i
1
i
1
r
4

+
r
1
|P
i
3
i
4
P
i
3
i
4
r
1
r
2
P
r
3
r
4
|i
1
]
3
p
2
r
3
r
4
i
1
p
2
i
3
i
4
r
1
p
2
i
3
i
4
r
1
r
2
[r
3
r
4
][r
4
i
1
] i
3
i
4
i
4
r
1
i
2
|P
r
4
i
1
|r
3
] i
3
|P
i
4
r
1
|r
2
]i
2
|P
i
3
i
4
r
1
|r
2
]
+ symmetry ip. (81)
This is equivalent to the gauge theory expression [13, 2, 29].
For (5, 3), the four possible step diagrams are drawn in Figure 5.
a) The rst diagram in Figure 5 corresponds to a residue evaluated on the poles f
11
= f
21
=
f
31
= 0. This diagram has p = 0, with n
0
= n = 3, m
1
= 4 and S
1
= {r
3
, i
1
, i
2
, i
3
, i
4
}. From
the general formula (80) the contribution to M
5,3
yields the rst term in (82).
b) The second diagram in Figure 5 corresponds to a residue evaluated at the poles f
11
=
f
32
= f
42
= 0. This diagram has p = 1, with n
0
= n = 3, m
1
= 2, n
1
= 2, m
2
= 4, and S
1
=
{r
3
, i
1
, i
2
}, S
2
= {i
5
, r
1
}, T
1
= {r
2
, r

3
, i

2
}, where P
T
1
= P
T

1
, with T

1
= {i
3
, i
4
, i
5
, r
1
},
and contributes the second term in (82).
Figure 5. Step Diagrams for the (5, 3) Split Helicity Amplitude.
The amplitude M
5,3
(i
1
, i
2
, i
3
, i
4
, i
5
, r
1
, r
2
, r
3
) remains invariant under the transformation
i
1
i
5
, i
2
i
4
, i
3
i
3
, r
1
r
3
, r
2
r
2
, and the remaining two step diagrams are re-
lated to the two we have computed above, by this transformation, the third to the second
23
and the fourth to the rst; so altogether we nd
M
5,3
=
[i
5
|P
r
1
r
2
|r
3

3
p
2
i
5
r
1
r
2
[i
5
r
1
][r
1
r
2
]r
3
i
1
i
1
i
2
i
2
i
3
i
3
i
4
[r
2
|P
i
5
r
1
|i
4

r
1
|P
i
3
i
4
i
5
P
i
1
i
2
|r
3

3
p
2
r
3
i
1
i
2
p
2
i
3
i
4
i
5
r
1
i
3
|P
i
4
i
5
r
1
|r
2
]i
2
|P
r
3
i
1
|r
2
] r
3
i
1
i
1
i
2
i
3
i
4
i
4
i
5
i
5
r
1

+ symmetry ip. (82)

This is equivalent to the gauge theory expression [26]-[28], [25].
8 Non-Split Helicity Tree Amplitudes
The methods developed here for calculating split helicity amplitudes can be extended to the
non-split helicity case. It is not so obvious how to choose the analogue of contiguous con-
straints for non-split helicity amplitudes but we can infer an appropriate choice by using a
procedure for deriving the non-split integrand from the split integrand. The integrand func-
tion

F
NS
mn
(c) for the tree amplitude (i
1
, . . . , i
a
, r
b
, r
b1
. . . r
1
, i
m
, . . . , i
a+2
, i
a+1
, r
b+1
, . . . , r
n
),
where, as usual, the indices i indicate positive helicities and the indices r negative helicities,
is related to the integrand function

F
mn
(c) for the split helicity case by

F
NS
mn
(c) =
(
ia

i
a+1
)(
r
b

r
b+1
)
(
ia

r
b
)(
i
a+1

r
b+1
)

F
mn
(c)
=
f
ab
c
a,b+1
c
a+1,b

F
mn
(c), (83)
where
k
denote the twistor string variables as described in [1]. By repeating this procedure
a number of times any non-split helicity amplitude can be obtained from a split helicity
amplitude. The procedure (83) does not introduce more poles into

F
NS
mn
(c), beyond the those
present in

F
mn
(c) as a result of the denominator f
11
f
m1,n1
of (13), for most values of a, b
unless m or n is small. For example if a = 1, b = 1, so that we are discussing an amplitude
with the helicity structure (+, , +, . . . , +, , . . . ), we see from (83) that the denominator
of

F
NS
mn
(c) is just f
m1,n1
, so that methods similar to those described in section 3 can be
used.
Acknowledgements
We are grateful to Nima Arkani-Hamed, Jacob Bourjaily, Freddy Cachazo, Dhritiman Nan-
dan, Mark Spradlin, Jaroslav Trnka, Anastasia Volovich and Cong-kao Wen for discussions.
LD thanks the Institute for Advanced Study at Princeton for its hospitality. LD was partially
supported by the U.S. Department of Energy, Grant No. DE-FG02-06ER-4141801, Task A.
24
A Evaluations of Jacobians
A.1 Residue Evaluation: the Jacobian

We consider the calculation of the Jacobian

f

(f
11
, f
12
, . . . , f
1,n2
)
(

22
,

23
, . . . ,

2,n1
)
, where f
1b
= c
1b
c
2,b+1
c
2b
c
1,b+1
. (84)
From (19),
c
1b
= a
i
1
r
b
+ [i
2
, i
3
]
_
r
b2
, r
b1

2,b1
r
b1
, r
b+1

2,b
+r
b+1
, r
b+2

2,b+1
_
c
2b
= a
i
2
r
b
[i
1
, i
3
]
_
r
b2
, r
b1

2,b1
r
b1
, r
b+1

2,b
+r
b+1
, r
b+2

2,b+1
_
+ . . . ,
where we have omitted terms in the last line that do not involve

2b
. Then, writing K
b
=
[i
1
, i
3
]c
1,b
+ [i
2
, i
3
]c
2,b
, the matrix
f

is given by an (n 2) (n 2) matrix beginning

_
_
_
_
_
_
_

12
, r
3

23
, r
1
K
4
r
1
, r
2
0 . . .
K
1
r
3
, r
4

23
, r
4

34
, r
2
K
5
r
2
, r
3
. . .
0 K
2
r
4
, r
5

34
, r
5

45
, r
3
. . .
0 0 K
3
r
5
, r
6

45
, r
6
. . .
.
.
.
.
.
.
.
.
.
.
.
.
_
_
_
_
_
_
_
and ending
_
_
_
_
_
_
_
.
.
.
.
.
.
.
.
.
.
.
.
. . .
n5,n4
, r
n3

n4,n3
, r
n4
K
n2
r
n5
, r
n4
0
. . . K
n5
r
n2
, r
n3

n4,n3
, r
n2

n3,n2
, r
n4
K
n1
r
n4
, r
n3

. . . 0 K
n4
r
n1
, r
n2

n3,n2
, r
n1

n2,n1
, r
n3

. . . 0 0 K
n3
r
n
, r
n1

n2,n1
, r
n

_
_
_
_
_
_
_
,
with
ab
= K
a

ra
+ K
a+1

r
a+1
+ . . . + K
b

r
b
. If the corresponding Jacobian determinant is
denoted by J
f
n
, this gives the recurrence relationship
J
f
n
=
n2,n1
, r
n
J
f
n1
K
n3
r
n
, r
n1

n2,n1
, r
n3
J
f
n2
+ K
n1
K
n3
K
n4
r
n
, r
n1
r
n1
, r
n2
r
n4
, r
n3
J
f
n3
, (85)
which will determine J
f
n
subject to
J
f
3
=
12
, r
3
, J
f
4
= K
2
r
2
, r
3

13
, r
4
, J
f
5
= K
2
K
3
r
2
, r
3
r
3
, r
4

14
, r
5
.
25
We now show by induction that
J
f
n
=
1,n1
, r
n

n2

b=2
K
b
r
b
, r
b+1
. (86)
If (86) holds, the right hand side of the recurrence relation (85) is
(
n2,n1
, r
n

1,n2
, r
n1
r
n3
,r
n2
r
n
, r
n1

n2,n1
, r
n3

1,n3
, r
n2

+K
n1
r
n
, r
n1
r
n1
, r
n2

1,n4
, r
n3
) K
n3
K
n4
r
n4
, r
n3

n5

b=2
K
b
r
b
, r
b+1
.
(87)
Now part of (87) is given by
(
n2,n1
, r
n

1,n4
, r
n1
r
n3
,r
n2
r
n
, r
n1

n2,n1
, r
n3

1,n4
, r
n2

+ K
n1
r
n
, r
n1
r
n1
, r
n2

1,n4
, r
n3
)
= K
n2
r
n3
, r
n2
(r
n2
, r
n

1,n4
, r
n1
+r
n
, r
n1

1,n4
, r
n2
)
= K
n2
r
n3
, r
n2
r
n2
, r
n1

1,n4
, r
n
.
The remaining terms in (87) are

n2,n1
, r
n

n3,n2
, r
n1
r
n3
, r
n2
r
n
, r
n1

n2,n1
, r
n3
K
n3
r
n3
, r
n2

= K
n2

n2,n1
, r
n
r
n2
, r
n1
r
n3
, r
n2

+
n2,n1
, r
n1
r
n3
, r
n
K
n3
r
n3
, r
n2

= K
n2

n3,n1
, r
n
r
n2
, r
n1
r
n3
, r
n2

so that (87) does indeed equal (86).

In order to compare the form of this Jacobian with terms coming from the residues in section
3, let f
1b
= 0, 1 b n 2, where
c
1b
i
1
, r
n

=
c
2b
i
2
, r
n

=
c
12
bn
i
1
, i
2

, K
b
=
r
n
, i
1
[i
1
, i
3
] +r
n
, i
2
[i
2
, i
3
]
r
n
, i
2

c
2b
=
r
n
|P
i
1
i
2
|i
3
]
r
n
, i
2

c
2b
,

1,n1
=
n1

b=1
K
b

r
b
=
r
n
|P
i
1
i
2
|i
3
]
r
n
, i
2

(
i
2
c
2n

rn
).
So

1,n1
, r
n
= r
n
|P
i
1
i
2
|i
3
],
26
and

= J
f
n
=
r
n
|P
i
1
i
2
|i
3
]
n2
r
n
, i
2

n3
n2

b=2
c
2b
r
b
, r
b+1

=
r
n
|P
i
1
i
2
|i
3
]
n2
i
1
, i
2
r
n
, i
1

r
n
, i
2

n1
n2

b=2
r
b
, r
b+1

_
1
c
11
f
1,n1
n1

b=1
c
2b
_
. (88)
A.2 Variable Change: the Jacobian

Had we not made the change of variables in the beginning of this section, it would have been
dicult to compute a general form for the Jacobian. However, we are not quite nished. If
we use the conditions f
1b
= 0, 1 b n 2, to eliminate
i
1
,
i
1
, c
1b
, 1 b n, to dene a
new system as in section 4(a), consisting of

i
2
,

i
2
=
1
i
2
, r
n

P
S

rn
;
ia
,
ia
, 3 a m; (89)

r
b
,
r
b
, 1 b n 1;
rn
,

rn
=
1
r
n
, i
2

P
S

i
2
; (90)
where S = {i
1
, i
2
, r
n
}, and c
ab
, 2 a m, 1 b n, and then use this system to dene
a set of contiguous s,

ab
, 3 a m 1, 2 b n 1, these will not be the same as
the corresponding

ab
. So we need to calculate the corresponding Jacobian. [Note that in
section 4.1, we used the constraints to eliminate
i
2
,
i
2
and redened

i
1
, but here we must
rephrase this because the remaining constraints involve c
2b
, 1 b n, and not c
1b
.]
To calculate the appropriate Jacobian, let
B
ab
= r
b2
, r
b1

a,b1
+r
b1
, r
b+1

ab
r
b+1
, r
b+2

a,b+1
, 3 b n 2,
B
a1
= r
2
, r
3

a2
, B
a2
= r
1
, r
3

a2
r
3
, r
4

a3
,
B
a,n1
= r
n3
, r
n2

a,n2
+r
n2
, r
n

a,n1
, B
an
= r
n2
, r
n1

a,n1
,
and similarly for B

ab
in terms of

ab
. Then
c
1b
= a
1b
[i
2
, i
3
]B
2b
, c
2b
= a
2b
+ [i
1
, i
3
]B
2b
[i
3
, i
4
]B
3b
,
c
3b
= a
3b
[i
1
, i
2
]B
2b
+ [i
2
, i
4
]B
3b
[i
4
, i
5
]B
4b
.
So, from c
1b
i
2
, r
n
= c
2b
i
1
, r
n
,
r
n
, i
1
[i
3
, i
4
]B
3b
= r
n
|P
i
1
i
2
|i
3
]B
2b
+r
n
, i
1
a
2b
r
n
, i
2
a
1b
c
2b
= a

2b
[i
3
, i
4
]B

3b
, c
3b
= a

3b
+ [i

1
, i
4
]B

3b
[i
4
, i
5
]B

4b
,
27
r
n
, i
1

r
n
, i
2

[i
3
, i
4
]B

3b
= [i
2
, i
3
]B
2b
+
r
n
, i
1

ir
n
, i
2

1b
a
1b
r
n
|P
i
1
i
2
|i
3
]B

3b
= [i
2
, i
3
]r
n
, i
2
B
3b
+ . . .
from which it follows that

3b
=
[i
2
, i
3
]r
n
, i
1

r
n
|P
i
1
i
2
|i
3
]

3b
+
3b
,

ab
=
ab
+
ab
, 4 a m1, 2 b n 1,
where
ab
depends on the momenta and the
a

b
, a

< a, and, hence,

=
_
[i
2
, i
3
]r
n
, i
2

r
n
|P
i
1
i
2
|i
3
]
_
n2
. (91)
28
References
[1] L. Dolan and P. Goddard, Gluon Tree Amplitudes in Open Twistor String Theory,JHEP
0912, 032 (2009)arXiv:0909.0499 [hep-th]].
[2] R. Britto, F. Cachazo and B. Feng, New recursion relations for tree amplitudes of gluons,
Nucl. Phys. B 715, 499 (2005) [arXiv:hep-th/0412308].
[3] R. Britto, F. Cachazo, B. Feng and E. Witten, Direct proof of tree-level recursion relation
in Yang-Mills theory, Phys. Rev. Lett. 94, 181602 (2005) [arXiv:hep-th/0501052].
[4] N. Arkani-Hamed, F. Cachazo, C. Cheung and J. Kaplan, The S-Matrix in twistor space,
arXiv:0903.2110 [hep-th].
[5] N. Arkani-Hamed, F. Cachazo, C. Cheung and J. Kaplan, A Duality for the S Matrix,
arXiv:0907.5418 [hep-th].
[6] N. Arkani-Hamed, F. Cachazo and C. Cheung, The Grassmannian origin of dual super-
conformal invariance, arXiv:0909.0483 [hep-th].
[7] N. Arkani-Hamed, J. Bourjaily, F. Cachazo and J. Trnka, Local spacetime physics from
the Grassmannian, arXiv:0912.3249 [hep-th].
[8] N. Arkani-Hamed, J. Bourjaily, F. Cachazo and J. Trnka, Unication of residues and
Grassmannian dualities, arXiv:0912.4912 [hep-th].
[9] E. Witten, Perturbative gauge theory as a string theory in twistor space, Commun. Math.
Phys. 252, 189 (2004) [arXiv:hep-th/0312171].
[10] N. Berkovits, An alternative string theory in twistor space for N = 4 super-Yang-Mills,
Phys. Rev. Lett. 93, 011601 (2004) [arXiv:hep-th/0402045].
[11] N. Berkovits and E. Witten, Conformal supergravity in twistor-string theory, JHEP
0408, 009 (2004) [arXiv:hep-th/0406051].
[12] N. Berkovits and L. Motl, Cubic twistorial string eld theory, JHEP 0404, 056 (2004)
[arXiv:hep-th/0403187].
[13] R. Roiban, M. Spradlin and A. Volovich, Dissolving N = 4 loop amplitudes into QCD
tree amplitudes, Phys. Rev. Lett. 94, 102002 (2005) [arXiv:hep-th/0412265].
[14] R. Roiban, M. Spradlin and A. Volovich, A googly amplitude from the B-model in twistor
space, JHEP 0404, 012 (2004) [arXiv:hep-th/0402016].
[15] R. Roiban and A. Volovich, All googly amplitudes from the B-model in twistor space,
Phys. Rev. Lett. 93, 131602 (2004) [arXiv:hep-th/0402121].
[16] R. Roiban, M. Spradlin and A. Volovich, On the tree-level S-matrix of Yang-Mills theory,
Phys. Rev. D 70, 026009 (2004) [arXiv:hep-th/0403190].
[17] F. Cachazo, P. Svrcek and E. Witten, MHV vertices and tree amplitudes in gauge theory,
JHEP 0409, 006 (2004) [arXiv:hep-th/0403047].
[18] M. L. Mangano and S. J. Parke, Multiparton amplitudes in gauge theories, Phys. Rept.
200, 301 (1991) [arXiv:hep-th/0509223].
[19] S. J. Parke and T. R. Taylor, An Amplitude for n Gluon Scattering, Phys. Rev. Lett.
56, 2459 (1986).
29
[20] F. A. Berends, W. T. Giele and H. Kuijf, On relations between multi - gluon and multi-
graviton scattering, Phys. Lett. B 211, 91 (1988).
[21] R. Penrose, Twistor algebra, J. Math. Phys. 8 345 (1967).
[22] V.P. Nair, A current algebra for some gauge theory amplitudes, Phys. Lett. 214 215
(1988).
[23] M. Spradlin and A. Volovich, From twistor string theory to recursion relations, Phys.
Rev. D 80, 085022 (2009), arXiv:0909.0229 [hep-th].
[24] D. Nandan, A. Volovich and C. Wen, A Grassmannian etude in NMHV minors,
arXiv:0912.3705 [hep-th].
[25] R. Britto, B. Feng, R. Roiban, M. Spradlin and A. Volovich, All split helicity tree-level
gluon amplitudes, Phys. Rev. D 71, 105017 (2005) [arXiv:hep-th/0503198].
[26] D. A. Kosower, Light cone recurrence relations for QCD amplitudes, Nucl. Phys. B 335,
23 (1990).
[27] D. A. Kosower, Next-to-maximal helicity violating amplitudes in gauge theory, Phys.
Rev. D 71, 045007 (2005) [arXiv:hep-th/0406175].
[28] M. x. Luo and C. k. Wen, Recursion relations for tree amplitudes in super gauge theories,
JHEP 0503, 004 (2005) [arXiv:hep-th/0501121].
[29] A. P. Hodges, Scattering amplitudes for eight gauge elds, arXiv:hep-th/0603101.
[30] A. Hodges, Eliminating spurious poles from gauge-theoretic amplitudes, arXiv:0905.1473
[hep-th].
[31] J. M. Drummond and J. M. Henn, All tree-level amplitudes in N=4 SYM, JHEP 0904,
018 (2009), arXiv:0808.2475 [hep-th].
[32] Z. Bern, L. J. Dixon and D. A. Kosower, All next-to-maximally helicity-violating one-
loop gluon amplitudes in /DHN = 4 super-Yang-Mills theory, Phys. Rev. D 72, 045014
(2005) [arXiv:hep-th/0412210].
[33] Z. Bern, V. Del Duca, L. J. Dixon and D. A. Kosower, All non-maximally-helicity-
violating one-loop seven-gluon amplitudes in N = 4 super-Yang-Mills theory, Phys. Rev.
D 71, 045006 (2005) [arXiv:hep-th/0410224].
[34] L. Dolan and P. Goddard, Tree and loop amplitudes in open twistor string theory, JHEP
0706, 005 (2007) [arXiv:hep-th/0703054].
[35] L. Dolan and P. Goddard, Current algebra on the torus, Commun. Math. Phys. 285,
219 (2009) [arXiv:0710.3743 [hep-th]].
[36] W. Fulton and J. Harris, Representation theory: a rst course, Springer: New York,
2004.
30